Blog · De Novo
FDA De Novo AI device product codes: a complete map
Every De Novo grant that created a new AI product code, and what its special controls mean for the devices clearing into it.
Every busy AI device product code started somewhere. In most cases, it started with a De Novo grant — FDA's decision to authorize a genuinely novel device type that had no predicate, simultaneously creating the regulatory category that subsequent devices would use to clear via 510(k).
How De Novo creates product codes
When a De Novo grant creates a new product code, it does three things with regulatory consequence: classifies the device type (almost always Class II for AI diagnostic software), establishes the special controls, and creates the predicate that subsequent manufacturers can cite in 510(k) submissions.
This is why De Novo grants are strategically significant beyond the individual device they authorize. The first company to clear a genuinely novel AI indication often defines the regulatory standards that competitors must meet.
How to read De Novo special controls
The special controls in a De Novo order define what manufacturers must demonstrate to clear a 510(k) in that product code. For AI devices, special controls typically cover algorithm performance thresholds, software documentation requirements, labeling requirements, and post-market obligations.
Reading the De Novo order closely — particularly the special controls section — tells you what FDA will expect from your 510(k) in that product code. This is more useful regulatory intelligence than any general guidance document on AI devices, because it's the specific standard applied to a specific indication.
De Novo timing and competitive dynamics
FDA's median De Novo review time has historically been 12 to 18 months, compared to 3 to 6 months for a 510(k). The submission requires more extensive safety and effectiveness documentation. The De Novo pathway becomes relevant for genuinely novel device types with no cleared predicate, or where a new regulatory framework would be advantageous over adapting to an existing category.